EP1940041A1

EP1940041A1 - Baseband processing method for improving signal-to-noise ratio based on multiple sampling

Info

Publication number: EP1940041A1
Application number: EP05785082A
Authority: EP
Inventors: Hai Jiang; Ping Li; Peng Geng
Original assignee: ZTE Corp
Current assignee: ZTE Corp
Priority date: 2005-09-15
Filing date: 2005-09-15
Publication date: 2008-07-02
Anticipated expiration: 2025-09-15
Also published as: WO2007030972A1; EP1940041B1; KR20080043884A; CN101167262A; KR101041009B1; CN101167262B; EP1940041A4; HK1108523A1

Abstract

The present invention discloses a baseband processing method for improving signal-to-noise ratio (SNR) based on multiple sampling. The method includes the following steps: (a) Sampling is performed for N times in each chip for the received signals to obtain N-path sampled data; (b) channel estimation is performed for the received sampled signal of each path respectively; (c) the channel estimation of each path is compared with each other to determine the channel estimation that has relatively greater power as well as less noise correlation, thus constituting a matched filtering matrix; (d) joint detection is performed for users signals by using system matrix and matched filtering matrix, thus obtaining the symbol of each user. Said method of the present invention can improve SNR and system performance by selecting the channel estimation tap that has relatively greater power as well as uncorrelated noise in the multi-path samples and participating in the matched filtering interference elimination of the joint detection. It can be used in either base station or terminal, and can be used in either single antenna environment or multiple-antenna environment.

Description

Technical Field

The present invention refers to a baseband processing method for improving signal-to-noise ration (SNR) based on multiple sampling in a mobile communication system.

Technology Background

In a digital communication system, it is common to use shaping filter at the receiving/sending ends in order to limit communication bandwidth. In the case of using a shaping filter, the sampling time points at the receiving end will affect the SNR of the sampled signals, and the SNR will be the maximum by only sampling at certain time points. However, because of the randomness of wireless channels, the user multi-path delay changes randomly, which results in a random change of SNR of the sampled signals, as well as a certain loss of the system performance.
A method of increasing sampling speed is usually employed in order to improve the SNR of sampling points in random wireless channels. China patent with application number 00105846.0 and titled as "A signal processing method for terminal equipment side in CDMA communication system" proposes a signal processing method of multiple sampling by using joint detection which is applied in CDMA system. The principle for it is that the receiving end performs multi-path sampling within one chip and channel estimation is performed for each path respectively and then all paths participate in joint detection. The advantage is that under the condition of random multi-path delay, the average sampled signal SNR obtained by multi-path sampling is higher than the SNR of single path sampling, thus the system performance is improved.
However, because of the usage of a shaping filter at the receiving end, the multi-path noises of multi-path sampled signals within one chip are correlated, and the diversity gain brought on by the combination of multi-path signals is not the gain of MRC (maximum ration combining). In fact, instead of the sum of the SNR of multi-path signals (MRC), the SNR after the combination of multi-path signals is between the maximum SNR and the minimum SNR for the paths. In other words, the SNR after the combination of multi-path signals is less than the maximum SNR that could be obtained. Simulation also proves this, when the wireless channel is configured as synchronous and thus it is ensured that the sampling point of each multi-path of each user has the maximum SNR, the performance of multiple sampling based on the above process is worse than that of single sampling.

Summary of the Invention

The technical problem that needs to be solved in present invention is to propose a new baseband processing solution for improving signal-to-noise ratio (SNR) based on multiple sampling in order to resist the deteriorated SNR brought by the random wireless channel and to further improve SNR.
In order to solve the above technical problem, the idea of present invention is as follows: performing multi-path sampling in each individual chip of received signals; performing channel estimation for each path signal respectively; comparing the multi-path channel estimation of each user with each other and taking the channel estimation tap that has relatively greater power and less noise correlation to participate in the matched filtering in the joint detection.
In order to realize the idea of present invention, said invention provides a baseband processing method for improving signal-to-noise ratio (SNR) based on multiple sampling, which comprises the following steps:

(a) Sampling is performed for N times in each chip for the received signals to obtain sampled data of N paths: $e_{m}^{s}, s = 1, \dots, N;$
(b) A channel estimation ĥ^s,s = 1,...,N is performed for the received sampled signals of each path respectively;
(c) the channel estimation ĥ^s of each path is compared with each other to determine the channel estimation ${\overline{h}}_{MF}^{s}$
that has relatively greater power as well as less noise correlation, thus constituting a matched filtering matrix $A_{MF}^{s};$
;
(d) a joint detection is performed for the signals of users by using system matrix A^s and matched filtering matrix $A_{MF}^{s},$
thus obtaining the symbol d̂ of each user.

Furthermore, in step (b), said channel estimation ĥ^s is obtained by using the equation: ${\overline{h}}^{s} = M^{- 1} e_{m}^{s},$
in which $e_{m}^{s}$
is the training sequence signals of the s^th sampling, M is the launched training sequence matrix, ĥ^s is the channel impulse response of all the users in the local cell of the s^th path sampling.
Furthermore, the channel estimation ${\overline{h}}_{MF}^{s}$
in step (c) can be obtained by employing the following steps:

(c1) performing post processing for the channel estimation ĥ^s of each path and obtaining the tap position of user signal path, as well as the channel estimation ${\overline{h}}_{p}^{s} (i),$
i=1,...,N_m,s=1,...,N after the post processing, in which N_m is the length of the channel estimation ĥ^s of each path;
(c2) combining the channel estimation ${\overline{h}}_{p}^{s} (i)$
into a total channel estimation ${\overline{h}}^{total} (l), {\overline{h}}^{total} (l) |_{l = ((i - 1) N + s)} = {\overline{h}}_{p}^{s} (i),$
i=1,...,N_m ,s=1,...,N, according to the time order;
(c3) performing a descending sorting on |ĥ^total (l)| and obtaining the channel estimation sequence ĥ^total (k_l ),l=1,...,N·N_m after the sorting and then performing the following calculations in an order from k ₁ to k_N·NM : if |ĥ^total (k_l )| is greater than zero, then the tap of this channel impulse response in channel estimation sequence ĥ^total (l) before sorting is retained and other values of this tap within the chip just on the left and the chip just on the right in ĥ^total (l) are set as zeros;
(c4) reducing the total channel estimation sequence ĥ^total (l) which is obtained after the processing of step (c3) to the channel estimation ${\overline{h}}_{MF}^{s}$
of N-path sampling.

Furthermore, in step (c), channel estimation ${\overline{h}}_{MF}^{s}$
can also be realized by employing the following steps:

(c11) performing post processing for the channel estimation of each path and obtaining the tap position of user signal path, as well as the channel estimation ${\overline{h}}_{p}^{s} (i),$
i=1,...,N_m,s=1,...,N after the post processing, in which N_m is the length of the channel estimation ĥ^s of each path;
(c12) comparing the modulus of the channel estimation of N-path sampling within one chip time length; selecting the maximum value and setting the others as zeros to obtain the channel estimation ${\overline{h}}_{MF}^{s}$
of N-path sampling.

It can be seen from above that the method of the present invention participates in the matched filtering interference elimination of the joint detection by selecting the channel estimation tap that has relatively greater power as well as uncorrelated noise in multiple sampling, therefore improving the SNR and the system performance. The present invention is applicable to any mobile digital communication system, especially to the system that employs joint detection solution. It can be used in either base stations or terminals, and can be used in either the single antenna environments or multiple-antenna environments.

Brief Description of Drawings

Figure 1 is a diagram of the processing method for the first embodiment of the present invention;
Figure 2 is a diagram of the processing method for the second embodiment of present invention.

Best Embodiments for Carrying out the Present Invention

The two embodiments of the present invention will be further described in detail with reference to the attached figures.

The first example:

As shown in the figure 1, a baseband processing method for improving signal-to-noise ratio (SNR) based on multiple sampling employs the following steps:
Step 101: performing multi-path sampling in each chip for received signals;
It is assumed that N times of samplings with the same time interval are performed in each chip of received signals, and the training sequence signal of the s^th sampling is represented as $e_{m}^{s},$
s=1,...,N.
Step 102: performing channel estimation for the sampled signals of N paths respectively;
The received training sequence is obtained by multi-path superposition of the multi-user training sequence signals and the superposition with the noise interference afterwards, which can be represented as: $e_{m}^{s} = M h^{s} + n_{m}^{s}, s = 1, \dots, N$
wherein M is the launched training sequence matrix, ĥ^s is the channel impulse response of all the users in the local cell of the s^th path sampling, $n_{m}^{s}$
is the noise that is superimposed on the training sequence of the s^th path sampling, therefore the channel estimation ĥ^s of each path sampling can be represented as follows: ${\overline{h}}^{s} = M^{- 1} e_{m}^{s}, s = 1, \dots, N$
wherein the dimension of the channel impulse response ĥ^s of the s^th path sampling is assumed to be N_m × 1.
Step 103: performing post processing for the channel estimation ĥ of each path and obtaining the tap position of user signal path;
In other words, the power of each tap in ĥ^s is compared with a threshold; the tap that is lower than the threshold is considered as the noise, and its value is set as zero; the tap that is higher than the threshold is considered as the signal, and its value keeps unchanged. The channel estimation is assumed to be ${\overline{h}}_{p}^{s} (i),$
i=1,...,N_m,s=1,...,N after the post processing.
Step 104: Sorting the channel estimation ${\overline{h}}_{p}^{s,},$
s=1,...,N of the N-path sampling in time order and combining them into a total channel estimation ĥ^total (l), l=1,..., N·N_m, then: ${\overline{h}}^{total} (l) |_{l = ((i - 1) N + s)} = {\overline{h}}_{p}^{s} (i), i = 1, \dots, N_{m}, s = 1, \dots, N$
For example, assuming N_m is equal to 3, N is equal to 2, the channel estimation sequence ĥ ¹ is ĥ ¹(1),ĥ ¹(2),ĥ ¹(3), the channel estimation sequence ĥ ² is ĥ ²(1),ĥ ²(2),ĥ ²(3), then the sequence after combination is ĥ^total (l) = ĥ ¹(1),ĥ ²(1),ĥ ¹(2),ĥ ²(2),ĥ ¹(3),ĥ ²(3).
Step 105: processing the total channel estimation ĥ^total and selecting the tap that has relatively greater power and uncorrelated noises, setting other taps as zeros;
The sequence ĥ ^total(l), l=1,...,N·N_m is resorted in a descending order based on the value of |ĥ^total (k_l )|, and the sequence ĥ^total (k_l ), l=1,...,N·N_m , is obtained, i.e., |ĥ^total (k ₁)|≥|ĥ^total (k₂ )|≥···≥|ĥ^total (k _{N·N_m})|.
For k ₁, if |ĥ^total (k ₁)|>0, then: ${\overline{h}}^{total} (l) = 0, l = k_{1} - (N - 1), k_{1} - (N - 2), \dots, k_{1} - 1, k_{1} + 1, k_{1} + 2, k_{1} + (N - 1),$
ĥ^total (l) is the total channel estimation sequence before sorting. The operation on k ₁ is repeated for k ₂,k ₃,...,k _{N·N_m} , i.e., the tap that has relatively greater power and uncorrelated noises can be selected.
The process can be described as follows by using the algorithm language:
for l=1:N·N_m
if |ĥ ^total (k₁ )|>0
${\overline{h}}^{total} (l) = 0, l = k_{1} - (N - 1), k_{1} - (N - 2), \dots, k_{1} - 1, k_{1} + 1, k_{1} + 2, k_{1} + (N - 1)$

end if
end for
This step assumes that the noises of multi-path sampling within one chip are correlated, as a result, while retaining the channel impulse response tap that has relatively greater power, other values of this tap within the chip just on the left and the chip just on the right are set as zeros, and these taps do not participate in the matched filtering due to the reason that the combination of the noise-related sampled signals of each path will not bring about the gain.
Step 106: reducing ĥ^total (l) which is obtained after the process in step 105 into the channel estimation ${\overline{h}}_{MF}^{s}$
of N-path sampling, in which, ${\overline{h}}_{MF}^{s} (i) = {\overline{h}}^{total} ((i - 1) N + s), i = 1, \dots, N_{m}, s = 1, \dots, N$
Step 107: generating the scrambling code and spreading code of the local cell.
Step 108: using each user's scrambling code, spreading code and channel estimation ĥ^s to constitute a system matrix A^s .
Said step 107 and 108 could be executed in parallel with steps 103 to 106 after step 102 is completed, and they have no order relationships with the execution of steps 103 to 106.
Step 109: using each user's scrambling code, spreading code and the channel estimation ${\overline{h}}_{MF}^{s}$
of the matched filtering in joint detection to constitute the matched filtering matrix $A_{MF}^{s} .$
.
Step 110: performing joint detection for the user signals to obtain each user symbol d̂;
The multi-path sampled signal $e_{d}^{s}$
of the received data part could be represented as the following: $e_{d}^{s} = A^{s} d + n_{d}^{s}, s = 1, \dots, N$
Wherein A^s is the system matrix of the s^th path sampling, which is composed of each user's scrambling code, spreading code and channel estimation ĥ^s .
The matched filtering system matrix $A_{MF}^{s}$
of the s^th path sampling is composed of each user's scrambling code, spreading code and the channel estimation ${\overline{h}}_{MF}^{s}$
that participates in matched filtering. The estimation for user symbol d in (6) is as follows:

Zero Forcing Block Linear Equalizer: $\overline{d} = {(\sum_{s = 1}^{N} {(A_{MF}^{s})}^{H} A^{s})}^{- 1} \sum_{s = 1}^{N} {(A_{MF}^{s})}^{H} e_{d}^{s}$
Minimum Mean Square Error Block Linear Equalizer: $\overline{d} = {(\sum_{s = 1}^{N} {(A_{MF}^{s})}^{H} A^{s} + σ_{n}^{2} I)}^{- 1} \sum_{s = 1}^{N} {(A_{MF}^{s})}^{H} e_{d}^{s}$
wherein, $σ_{n}^{}$
is the noise power.

The key point in the method of present embodiment is to pick out the multi-path that has relatively greater power and uncorrelated noise from the multi-path sampled signals to participate in the matched filtering in joint detection, thus obtaining diversity gain.

The second example:

The method in the first embodiment needs to perform sorting for channel estimation of each path, which is complicated. The present embodiment simplifies it with the cost of a certain loss in performance, as shown in figure 2, the detailed description is as follows:
Step 201: performing chip multi-path sampling for the received signals;
It is assumed to perform N times of samplings with the same time interval in each chip for the received signals; the training sequence signal of the s^th sampling is represented as $e_{m}^{s},$
s=1,...,N.
Step 202: performing channel estimation for the sampled signals of N paths respectively;
The received training sequence is obtained by multi-path superposition of the multi-user training sequence signals and the superposition with the noise interference afterwards, which could be represented as: $e_{m}^{s} = M h^{s} + n_{m}^{s}, s = 1, \dots, N$
wherein M is the launched training sequence matrix, ĥ^s is the channel impulse response of all the users in the local cell of the s^th path sampling, $n_{m}^{s}$
is the noise that is superimposed on the training sequence of the s^th path sampling, therefore the channel estimation ĥ^s of each path sampling can be represented as follows: ${\overline{h}}^{s} = M^{- 1} e_{m}^{s}, s = 1, \dots, N$
wherein the dimension of the channel impulse response ĥ^s of the s^th path sampling is assumed to be N_m ×1.
Step 203: performing post processing for the channel estimation ĥ^s of each path and obtaining the tap position of user signal path;
In other words, the power of each tap in ĥ^s is compared with a threshold; the tap that is lower than the threshold is considered as the noise, with the set value as zero; the tap that is higher than the threshold is considered as the signal with its value unchanged. The channel estimation is assumed to be ${\overline{h}}_{p}^{s} (i),$
i = 1,...,N_m,s=1,...,N after the post processing.
Step 204, comparing the modulus of the channel estimation of each path sampling within a chip time length, picking the maximum value and setting others as zeros;
The process could be described as follows by using the algorithm language:
for i=1:N_m
in |h_p ^s (i)|, s=1,..., N with N values, selecting the maximum value and setting the sequence number as s _max, also setting: ${\overline{h}}_{MF}^{S_{\max}} (i) = {\overline{h}}^{s_{\max}} (i)$
${\overline{h}}_{MF}^{s} (i) = 0, s = 1, \dots, s_{\max} - 1, s_{\max} + 1, \dots, N$

end for.
Step 205: Generating the scrambling code and spreading code of the local cell.
Step 206: using each user's scrambling code, spreading code and channel estimation ĥ^s to constitute system matrix A^s .
Said step 205 and 206 could be executed in parallel with step 203 and step 204 after step 202 is completed, and they have no order relationships with the execution of step 203 and step 204.
Step 207: using each user's scrambling code, spreading code and the channel estimation ${\overline{h}}_{MF}^{s}$
of the matched filtering in joint detection to constitute the matched filtering matrix $A_{MF}^{s} .$
Step 208: performing joint detection for the user signals to obtain each user symbol d̂;
The multi-path sampled signal $e_{d}^{s}$
of the received data part can be represented as follows: $e_{d}^{s} = A^{s} d + n_{d}^{s}, s = 1, \dots, N$
Wherein A^s is the system matrix of the s^th path sampling, which is composed of each user's scrambling code, spreading code and channel estimation ĥ^s .
While at the same time, the matched filtering system matrix $A_{MF}^{s}$
is constructed, which is composed of each user's scrambling code, spreading code and the channel estimation ${\overline{h}}_{MF}^{s}$
that participates in matched filtering. The estimation on user symbol d in (12) is as follows:

Zero Forcing Block Linear Equalizer: $\overline{d} = {(\sum_{s = 1}^{N} {(A_{MF}^{s})}^{H} A^{s})}^{- 1} \sum_{s = 1}^{N} {(A_{MF}^{s})}^{H} e_{d}^{s}$
Minimum Mean Square Error Block Linear Equalizer: $\overline{d} = {(\sum_{s = 1}^{N} {(A_{MF}^{s})}^{H} A^{s} + σ_{n}^{2} I)}^{- 1} \sum_{s = 1}^{N} {(A_{MF}^{s})}^{H} e_{d}^{s}$
wherein $σ_{n}^{}$
is the noise power.

The key point in the method of the second embodiment is to compare the modulus of the channel estimation of N-path sampling within a chip time length, pick out the maximum value to participate in the matched filtering in joint detection and set other values as zeros without participating in the matched filtering in joint detection. This is simpler than the first solution. However, in the second solution, the noises between the taps that have the maximum channel estimation power in each chip may still be correlated, thus the performance may be lost to a certain degree.

Industry Applicability

Said solutions of the present invention improve the SNR and the system performance by selecting the channel estimation tap that has relatively greater power as well as uncorrelated noise in multiple sampling and participating in the matched filtering interference elimination of the joint detection. It can be used in either base stations or terminals, and can be used in the either single antenna environments or multiple-antenna environments.

Claims

A baseband processing method for improving signal-to-noise ratio (SNR) based on multiple sampling, comprising the following steps:
(a) sampling for N times within each chip for received signals to obtain sampled data of N paths: $e_{m}^{s},$
s=1,...,N;

(b) performing channel estimation ĥ^s,s=1,...,N for received sampled signals of each path respectively;

(c) comparing the channel estimation ĥ^s of each path to determine a channel estimation ${\overline{h}}_{MF}^{s}$
that has relatively greater power as well as less noise correlation, thus constituting a matched filtering matrix $A_{MF}^{s};$

(d) performing a joint detection for users signals by using a system matrix A^s and the matched filtering matrix $A_{MF}^{s},$
thus obtaining symbol d̂ of each user.
The baseband processing method for improving signal-to-noise ratio (SNR) based on multiple sampling of claim 1, characterized in that in step (b), said channel estimation ĥ^s is obtained by using an equation: ${\overline{h}}^{s} = M^{- 1} e_{m}^{s},$
wherein $e_{m}^{s}$
is a training sequence signal of an s^th sampling, M is a launched training sequence matrix, ĥ^s is channel impulse response of all the users in local cell of the s^th path sampling.
The baseband processing method for improving signal-to-noise ratio (SNR) based on multiple sampling of claim 1, wherein the channel estimation ${\overline{h}}_{MF}^{s}$
in step (c) can be obtained by employing the following steps:
(c1) performing post processing for the channel estimation ĥ^s of each path and obtaining a tap position of a user signal path, as well as a channel estimation ${\overline{h}}_{p}^{s} (i),$
i=1,...,N_m,s=1,...,N after the post processing, wherein N_m is a length of the channel estimation ĥ^s of each path;

(c2) combining the channel estimation ${\overline{h}}_{p}^{s} (i)$
in time order into a total channel estimation ${\overline{h}}^{total} (l), {\overline{h}}^{total} (l) |_{l = ((i - 1) N + s)} = {\overline{h}}_{p}^{s} (i),$
i=1,...,N_m ,s=1,...,N,;

(c3) performing a descending sorting on |ĥ^total (l)| and obtaining a channel estimation sequence ĥ^total (k_l ),l=1,...,N·N_m after the sorting and then performing following calculations in an order starting from k ₁ to k_N·NM : if |ĥ^total (k_l )| is greater than zero, then an impulse response tap of this channel in channel estimation sequence ĥ^total (l) before sorting is retained and other values of this tap within a chip just on the left and a chip just on the right in ĥ^total (l) are set as zeros;

(c4) reducing the total channel estimation sequence ĥ^total (l) which is obtained after the processing of step (c3) to the channel estimation ${\overline{h}}_{MF}^{s}$
of N-path sampling.
The baseband processing method for improving signal-to-noise ratio (SNR) based on multiple sampling of claim 1, wherein in step (c), the channel estimation ${\overline{h}}_{MF}^{s}$
can also be realized by employing the following steps:
(c11) performing post processing for the channel estimation of each path and obtaining the tap position of the user signal path, as well as the channel estimation ${\overline{h}}_{p}^{s} (i),$
i =1,...,N_m,s =1,...,N after the post processing, wherein N_m is the length of the channel estimation ĥ^s of each path;

(c12) comparing a modulus of the channel estimation of N-path sampling within a chip time length; selecting a maximum value and setting others as zeros to obtain the channel estimation ${\overline{h}}_{MF}^{s}$
of N-path sampling.
The baseband processing method for improving signal-to-noise ratio (SNR) based on multiple sampling of claim 1, wherein in step (c1) and (c11), the channel estimation ${\overline{h}}_{p}^{s} (i)$
after said post processing is obtained as follows: comparing the power of each tap in ĥ^s with a threshold; the tap that is lower than the threshold is considered as noise, and its value is set as zero; the tap that is higher than the threshold is considered as a signal, and its value is retained.
The baseband processing method for improving signal-to-noise ratio (SNR) based on multiple sampling of claim 1, wherein in step (d), said system matrix A^s is composed of each user's spreading code, scrambling code and the channel estimation ĥ^s .
The baseband processing method for improving signal-to-noise ratio (SNR) based on multiple sampling of claim 1, wherein in step (d), a symbol d̂ of each said user is estimated based on a multi-path sampled signal $e_{d}^{s}$
of received data part, the system matrix A^s and the matched filtering matrix $A_{MF}^{s},$
and an equation is as follows: $\overline{d} = {(\sum_{s = 1}^{N} {(A_{MF}^{s})}^{H} A^{s})}^{- 1} \sum_{s = 1}^{N} {(A_{MF}^{s})}^{H} e_{d}^{s}$
or $\overline{d} = {(\sum_{s = 1}^{N} {(A_{MF}^{s})}^{H} A^{s} + σ_{n}^{2} I)}^{- 1} \sum_{s = 1}^{N} {(A_{MF}^{s})}^{H} e_{d}^{s},$
wherein $e_{d}^{s}$
is the multi-path sampled signal of the received data part, $e_{d}^{s} = A^{s} d + n_{d}^{s},$
s=1,..., N, $σ_{n}^{2}$
is noise power.